Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

Learning Disability Quarterly

36(4) 203 214


Hammill Institute on Disabilities 2012
Reprints and permissions:
sagepub.com/journalsPermissions.nav
DOI: 10.1177/0731948712464034
ldq.sagepub.com
Recent intervention studies directed to improve problem-
solving accuracy in children with math difficulties (MD)
have found support for teaching cognitive strategies (see
Powell, 2011, for a review). Several studies have found that
verbal strategy instruction (e.g., Montague, 2008; Montague,
Warger, & Morgan, 2000, Xin, 2008), as well as visual-
spatial strategies (e.g., Kolloffel, Eysink, de Jong, &
Wilhelm, 2009; van Garderen, 2007), enhance childrens
math performance relative to control conditions (see Baker,
Gersten, & Lee, 2002; Gersten et al., 2009, for reviews).
One set of strategies that has been overlooked in the word
problem solving literature, however, is text comprehension
strategies. This is unfortunate because reading comprehen-
sion has been found to be highly predictive of solution accu-
racy (e.g., Cornoldi, Drusi, Tencati, Giofr, & Mirandola,
2012; Kail & Hall, 1999; Swanson, Cooney, & Brock,
1993; Vilenius-Tuohimaa, Aunola, & Nurmi, 2008) and in
some cases is a better predictor of solution accuracy than
calculation skills (e.g., Swanson et al., 1993). One strategy
found effective in text comprehension is generative learn-
ing (e.g., Hooper, Sales, & Rysavy, 1994; Plass, Chun,
Mayer, & Leutner, 1998; Wittrock, 1989, 1991). Genera-
tive learning requires the student to paraphrase the text
either orally and/or in writing (other forms include draw-
ing a picture of the main idea, etc.) prior to responding to
questions about the text (e.g., Schwamborn, Mayer, Thillmann,
Leopold, & Leutner, 2010; Wittrock, 1991). Such genera-
tive activities are assumed to assist students in cognitive
and metacognitive processing and thus foster a deep under-
standing of the material (by mentally connecting the new
information with relevant prior knowledge) to be learned
(e.g., see Wittrock, 1989, for a review).
In this study, we investigate whether generative strategies
directed toward propositions within word problems enhance
solution accuracy in children with MD. We assumed that
because children with MD experience difficulties in prob-
lem representation (e.g., Passolunghi, Marzocchi, & Fiorillo,
2005), misunderstandings occur when they construct a men-
tal model of a problem that conflicts with information (the
propositions) in the problem statement. In the intervention
tested here, we assumed the generative strategies should
help students avoid this conflict by focusing attention on the
464034LDQ36410.1177/0731948712464034
Learning Disability QuarterlySwanson et al.
1
University of CaliforniaRiverside, USA
2
University of CaliforniaSanta Barbara, USA
Corresponding Author:
H. Lee Swanson, Department of Educational Psychology and Special
Education, Graduate School of Education, University of California,
1207 Sproul Hall, Riverside, CA 92521, USA
Email: lee.swanson@ucr.edu
Generative Strategies, Working Memory,
and Word Problem Solving Accuracy in
Children at Risk for Math Disabilities
H. Lee Swanson, PhD
1
, Amber S. Moran, PhD
2
, Kathleen Bocian, PhD
1
,
Cathy Lussier, PhD
1
, and Xinhua Zheng, PhD
1
Abstract
This study investigated the role of generative strategies and working memory capacity on word problem solving accuracy in
children with math difficulties (MD). Within classrooms, children in Grade 3 with MD (n = 69) were randomly assigned to
one of three treatment conditions: paraphrase question propositions (Restate), paraphrase relevant propositions (Relevant),
and paraphrase all propositions (Complete), or to an untreated control. An additional control group included children
without MD (n = 22). Mixed regression modeling showed that generative strategies significantly improved posttest scores
for children with MD compared with the control condition, but outcomes were related to the type of dependent measures.
The Relevant and Complete treatment conditions improved problem-solving accuracy, the Complete condition improved
problem component identification, and the Restate and Relevant conditions improved operation span performance when
compared with the control conditions. Only the Relevant and Complete generative learning treatments allowed children
with MD to catch up to children without MD, but the results were moderated by working memory capacity.
Keywords
math diffculties, strategy training, working memory
Article
204 Learning Disability Quarterly 36(4)
key propositions in the problem statement. To this end, the
current study addresses three questions.
Research Question 1: Do generative strategy con-
ditions that focus on isolated propositions or all
propositions within word problems facilitate solu-
tion accuracy when compared with the control
condition for children with MD?
In this study, children with MD were randomly assigned
to one of three generative strategy treatment conditions or
a control condition. The treatment conditions included the
experimenter reading word problems aloud to the students
who were then asked to paraphrase (orally state and write)
particular propositions (sentences) within the word prob-
lem before solving the problem. Consider the following
propositions within a word problem: Darren found
15 pinecones. He threw 5 pinecones back. Darren found
1 pinecone that was red. How many pinecones did Darren
keep? From this problem, a student can focus on several
explicit and implicit propositions, but four are critical in
our treatment analysis. They are the Question (How many
pinecones did Darren have left?), Goal (Find the differ-
ence between the pinecones Darren kept and the ones he
threw back), Relevant Numbers (two assignment proposi-
tions that lead to the correct solution, 15 and 5), and
Irrelevant Information (Darren found 1 pinecone that
was red).
The question to be answered is whether generative strat-
egies need to focus on all propositions or isolated proposi-
tions to increase solution accuracy. To this end, one
treatment condition in this study focused solely on the ques-
tion propositions: The student was asked to restate in his or
her own words what the problem seeks to answer. This
activity we assumed would allow students to bypass extra-
neous information within the word problem and apply the
correct algorithm for solution (e.g., see Cook & Rieser,
2005, for a research review of the guided question strategy).
The second treatment focused on identifying the question
proposition and the relevant number propositions needed to
solve the problem. We assumed that this condition would
minimize cognitive demands by focusing only on the prop-
ositions that included numbers relevant to a solution. The
third treatment condition directed children to paraphrase all
propositions. In this way, an assessment can be made as to
whether a focus on specific isolated propositions varies
from the baseline (focus on all propositions). Children with
MD in the treatment and control conditions were also com-
pared with a control condition that included children with-
out MD.
Research Question 2: Do generative strategies make
demands on working memory capacity (WMC)?
Assuming there are potential benefits of generative strat-
egies in remediating word problem solving difficulties for
some children, the use of generative strategies for other
children with MD may not always be advantageous. This
may be because generative processing does not automati-
cally guarantee deep understanding, or because the logistics
of paraphrasing text may prime extraneous cognitive pro-
cessing, thereby leaving fewer cognitive resources avail-
able for carrying out the processes needed to solve the
problem. Thus, from an aptitude-treatment perspective, not
all children with MD may be expected to benefit from this
generative training. Because a significant predictor of prob-
lems solving accuracy is working memory (e.g., Swanson
& Beebe-Frankenberger, 2004), and because children with
MD experience serious working memory difficulties (e.g.,
Geary, 2010; Swanson, Jerman, & Zheng, 2008), such poor
problem-solving skills in these children, plus their low
WMC, may mitigate the effectiveness of generative strate-
gies. Thus, we explored whether individual differences in
WMC play a major role in the posttest outcomes.
Research Question 3: Does generative strategy
instruction facilitate transfer?
In addition to exploring whether generative strategies
influenced posttest measures of problem-solving accuracy,
we also determined whether transfer occurred on related
measures. Two measures were selected: component
identification within word problems and operation span.
Component identification was selected because all lessons
required recognition of problem-solving propositions.
Operation span was selected because it required the com-
bining of words and numbers related to calculations that
increased in set size.
Method
Participants
In all, 91 third-grade children participated in this study.
Participation was based on parent approval. Forty-six were
boys and 45 were girls. Ethnic representation of the sample
was 29% Anglo, 68% Hispanic, 2% Asian, and 1% mixed
and/or Other (e.g., Anglo and Hispanic, Native American).
Data were collected during 20092010 for the first cohort
and 20102011 for the second cohort. Children were
selected from 12 third-grade classrooms across four ele-
mentary schools in two Southwest U.S. school districts.
Identification of at risk for MD. There is controversy over
the definition of math disabilities, especially as it applies to
identifying risk factors related to solving word problems.
Because not all children identified as at risk in the study had
diagnosed learning disabilities, we used the term at risk
Swanson et al. 205
for math difficulties (MD). We examined children who per-
formed in the lower 25th percentile on norm-referenced
word problem solving math tests. The 25th percentile cutoff
score on standardized achievement measures has been com-
monly used to identify children at risk (e.g., Fletcher et al.,
1989; Siegel & Ryan, 1989). It is also important to note that
our focus was on children with word problem solving dif-
ficulties and not calculation difficulties, per se. Students
were identified as having MD if they scored between the
35th and 90th percentiles on a measure of fluid intelligence
(Raven Colored Progressive Matrices Test; Raven, 1976)
and had a mean scaled score of 8 (90 standard score or 25th
percentile) or below from two standardized word problem
solving mathematics measures: the Test of Mathematical
AbilitiesSecond Edition (TOMA-2; Brown, Cronin, &
McIntire, 1994) or KeyMath (KEYM; Connolly, 1998).
Using these criteria, a total of 69 students were identified
with MD and 22 without MD. All students in the study
received classroom instruction in English. Twenty-six chil-
dren whose first language was Spanish but were considered
English proficient based on their California English Lan-
guage Development Test (CELDT) scores were in the
sampling. No significant difference was found in terms of
English language learner (ELL) status between the treat-
ment conditions,
2
(4, N = 91) = 3.42, p > .05, and ELL
children with MD were evenly distributed across treatment
conditions.
Design and Treatment Conditions
Random assignment. Children with MD were randomly
assigned to one of four conditions: restatement of the ques-
tion (Restate, n = 18), restatement of the question and rele-
vant propositions (Relevant, n = 18), restatement of the
question and relevant and irrelevant propositions (Com-
plete, n = 18), or a control condition (MD Control, n = 15).
A separate control condition included children not at risk
for MD (NMD Control, n = 22).
Common instructional classroom conditions. The treatment
instruction in word problem solving supplemented school-
wide instruction in mathematics concepts and computation,
taught by the classroom teacher. The district curriculum
was Macmillan/McGraw-Hills Mathematics: Concepts,
Skills, and Problem SolvingThird Grade (Altier, Day, &
Balka, 2009). All students received supplemental instruc-
tion (either with a tutor delivering the treatment condition
or with the classroom teacher for control conditions) for a
30-min period twice a week for a period of 10 weeks, for a
total of 20 lessons in all.
Common instruction across experimental conditions. All
experimental treatment conditions presented 20 lessons of
five word problems per lesson. Word problems were modi-
fied from the classroom text, Macmillan/McGraw-Hills
Mathematics (Altier et al., 2009), and modifications were lim-
ited to the addition of irrelevant sentences. The number of
irrelevant sentences (sentences that did not contribute to the
solution) increased over the course of the lessons, beginning
with zero irrelevant sentences in Lessons 1 through 5 and cul-
minating in four irrelevant sentences in Lessons 16 through 20.
Each child in the treatment conditions received a booklet
containing the 20 lessons, where student work was recorded.
Tutors followed a script in the presentation of word prob-
lems across four lesson phases: warm-up, modeling (one
problem), guided practice (one problem), and independent
practice (three problems). Treatment conditions were pre-
sented in small groups of two to four students.
Regardless of treatment condition, children began the
sessions with a warm-up phase. Children completed prob-
lems that required calculations to find missing numbers
(e.g., 9 + 2 = x; x + 1 = 6; x 5 = 1) and used geometric
shapes to complete a series of puzzles. The warm-up phase
lasted up to 5 min.
The modeling phase, which lasted approximately 5 min,
consisted of a tutor modeling the specific treatment prob-
lem-solving strategy. The tutor read the first word problem
aloud and then asked the group to verbally restate the ques-
tion (restate what the problem is asking). Depending on the
treatment condition, however, students verbalized the tar-
geted problem-solving components specific to the treatment
problem-solving strategy via explicit questioning and with
corrective feedback from the tutor.
For the third phase, guided practice, the tutor read the
question aloud to the children twice. Children were directed
to restate in their own words (without looking back at the
problem) the targeted propositions specific to their treat-
ment and then to record these in their workbooks. Children
then stated what they had written to the tutor and the group,
and corrective feedback was provided. This guided practice
phase for each session lasted approximately 5 to 6 min.
The independent practice phase of the lesson directed
children to solve three word problems, similar in difficulty
to the two previously modeled and guided practice word
problems. Children were directed to read the problems on
their own and to solve the problems just like we did. The
time allocation for independent practice phase was 15 min.
Child workbooks were assessed each session to determine
strategy implementation and problem-solving accuracy.
Conditions specific to treatment. The three treatment con-
ditions focused on problem translation by directing children
to paraphrase (via writing) propositions of word problems.
The first treatment condition, the Restate condition (n = 18),
targeted the recall of two propositions: the question and
goal. The steps for this condition during the modeling phase
required the child to read the problem, find the question,
restate the question orally to the tutor, write the question in
his or her own words, and then solve the problem.
206 Learning Disability Quarterly 36(4)
The second treatment condition, the Relevant condition
(n = 18), targeted the paraphrasing of three propositions; the
question, goal, and numbers. As above, the steps for this
condition during the modeling phase of the lesson plan
required the child to read the problem, restate the question,
and later solve the problem. However, after the child ver-
bally restated the question, the tutor asked, What other
information is important? The child then stated the impor-
tant information (sentences that included the necessary
numbers for solution).The tutor provided feedback by
restating the question and relevant propositions. Prior to
solving the problem, the child was then directed to write the
question and all the important information in his or her own
words.
The third treatment condition, the Complete condition
(n = 18), directed the children to restate the question as well
as relevant and irrelevant propositions. As in the other two
conditions, the steps for this condition during the modeling
phase of the lesson plan required the child to read the prob-
lem and restate the question, write the proposition in his or
her own words, and later solve the problem. However, after
a child verbally restated the question, the tutor asked him or
her to state the necessary information (sentences that
included the numbers) and the sentences that included the
unnecessary numbers. The tutor provided feedback by
restating the question and the relevant and irrelevant
propositions.
The control conditions included children with MD (MD
Control n = 15) and without MD (NMD Control n = 22)
from the same classrooms as the children in the treatment
conditions. These children received small-group instruction
from the classroom teacher for 30 min twice each week for
10 weeks.
Treatment fidelity. Each tutor received the scripted les-
sons and student booklets, and he or she practiced deliver-
ing lessons to other tutors to receive implementation fidelity
feedback. Additional assistance, direct coaching, and fol-
low-up were planned for any tutors who fell below 80% on
the fidelity of implementation rubric. The fidelity of imple-
mentation rubric covered all segments of the lesson
interventionwarm-up, modeling, guided practice, and
independent practiceand included the specific components
that were central to each treatment condition. Mean imple-
mentation fidelity was 96.6%.
Measures
The battery of group and individually administered tasks is
described below. Experimental tasks are described in more
detail than are published and standardized tasks. Tasks
were divided into classification, pretest-only (moderator
measures), and pretest/posttest measures. The Cronbachs
alphas for each measure varied from .60 to .98.
Fluid intelligence. Fluid intelligence was assessed by the
Raven Colored Progressive Matrices (Raven, 1976). Chil-
dren were required to circle the replacement piece that best
completed the patterns.
Word recognition. Word recognition was assessed by the
Reading subtest of the Wide Range Achievement Test
(WRAT-III; Wilkinson, 1993). The task provided a list of
words of increasing difficulty. The dependent measure was
the number of words read correctly.
Reading comprehension. Reading comprehension was
assessed by the Passage Comprehension subtest from the
Test of Reading Comprehension (TORC-III; Brown,
Hammill, & Weiderholt, 1995).
Arithmetic calculation. The Arithmetic subtests from
the WRAT-III and the Wechsler Individual Achievement
Test (WIAT; Psychological Corporation, 1992) were
administered.
Writing speed. To assess whether differences existed
between treatment groups in writing speed, four tests were
administered. Each child was asked to write the alphabet in
sequential order as quickly as possible (30 s were allotted).
Children were then asked to quickly write letters in a ran-
dom, nonsystematic order (30 s were allotted). The same
procedure was followed with numbers; students were first
asked to write numbers from 0 to 9 in order as quickly as
possible in a 30-s period and then were asked to write num-
bers as quickly as possible out of order within a 30-s period
(i.e., 3, 9, 4, and 7). Combining results for both alphabetic
and numeric measures of writing speed showed no signifi-
cant differences among treatment conditions for children
with MD (F < 1.0).
Working Memory
Several studies show that individual differences in working
memory span (referred to here as working memory capacity,
WMC) play a major role in problem-solving performance
(e.g., Swanson et al., 2008); therefore, we measured WMC
to determine its effect on solution accuracy as a function of
treatment conditions. The WMC tasks required children to
hold increasingly complex information in memory while
responding to a question about the task. For this study, two
WMC tasks were administered (Conceptual Span and
Sentence/Digit Span task) that followed this format. A sepa-
rate WMC task, referred to as Updating, was also adminis-
tered. The tasks and administration are described in detail in
Swanson and Beebe-Frankenberger (2004).
WMC scores. After random assignment to conditions,
children were divided for later analysis into relatively high-
and low-WMC subgroups. A median split based on the
composite working memory score (average z score of Con-
ceptual Span, Digit Sentence Span, and Updating) for the
total sample at pretest was used to determine children of
Swanson et al. 207
relatively low and high WMC. As expected, the proportion
of children with low WMC occurred in the Restate
(66.67%), Relevant (72.22%), Complete (61.11%), and MD
Control (60%) conditions when compared with the NMD
Control condition (22.73%),
2
(4, N = 91) = 12.84, p < .05.
For children with MD, no significant differences (ps > .50)
in the proportion of children high and low in WMC emerged
among the treatment conditions.
Pretest and Posttest
The tasks described below included two equivalent forms
that were counterbalanced for presentation order.
Word problem accuracy. Three measures were adminis-
tered to assess word problem solving ability: the Word
Problem subtests from the TOMA-2 (Brown et al., 1994),
the KEYM, and the Story Problem subtest from the Com-
prehensive Mathematical Abilities Test (CMAT; Hresko,
Schlieve, Herron, Swain, & Sherbenou, 2003). The order of
the subtest presentation was counterbalanced across pretest
and posttest. Subtests from each of these measures yielded
a scale score (M = 10, SD = 3). The technical manual for this
subtest reported adequate reliabilities (>.86) and moderate
correlations (rs > .50) with other math standardized tests
(e.g., the Stanford Diagnostic Mathematics Test). A com-
posite score based on the mean of all three subtests was cre-
ated as the outcome measure of problem-solving ability.
We divided our tasks into near-transfer and far-transfer
measures. The near-transfer tasks (defined as tasks that
matched the focus of intervention) assess correct identifica-
tion of problem-solving components. For the far-transfer
measure (tasks not directly related to the focus of treat-
ment), we assessed improvements in operation span.
Operation span measures involved holding and manipulat-
ing both verbal and numerical information, and we assumed
that generative strategy instruction might facilitate such
performance. The strategy conditions also included
increasing demands in sentence load; therefore, children
were gaining practice in controlled attention, a process that
underlies the executive component of WMC. We describe
each of these tasks next.
Mathematical word problem solving components. This
experimental test assessed the childs ability to identify
(retrieve) processing components of word problems
(Swanson & Beebe-Frankenberger, 2004). This measure is
closely related to the intervention instruction that directed
childrens attention to the components of word problems. In
this task, participants were read a word problem. Each word
problem had four sentences that included two assignment
propositions (relevant sentence), one relation, one question,
and an extraneous proposition (irrelevant sentence) related
to the solution. After the word problem was read, partici-
pants were instructed to turn the page (so they could not
look back at the problem). For each word problem,
participants were prompted to identify (from four possible
choices) the question, the numbers, what the question wants
you to find, how to solve the problem (addition, subtraction,
or multiplication), and the number sentence. At the end of
the book, students were read a series of true/false state-
ments related to irrelevant information. The total score pos-
sible for propositions related to question, number, goal,
operations, algorithms, and truefalse questions was 12.
Operation span. A version of the Turley-Ames and Whit-
field (2003) Operation Span task, modified for children
(Swanson, Kehler, & Jerman, 2010), was individually
administered at pretest and posttest. The Operation Span
test assessed WMC span by having participants solve sim-
ple math problems while remembering unrelated to-
be-remembered words that follow each math problem.
Operation-word sequences increased in set size.
Statistical Analysis
Because the data reflected treatments of students nested
within classrooms, a hierarchical linear model (HLM; Bryk
& Raudenbush, 2002) was used to analyze treatment effects
nested within classrooms.
1
The fixed- and random-effect
parameter estimates were obtained using PROC MIXED in
SAS 9.3 (SAS Institute, 2010). The criterion variables in
the analysis were posttest scores for solution accuracy, cor-
rect identification of problem-solving components, and
operation span.
Results
Descriptive Analysis
Table 1 presents the demographic data of children at pretest
by treatment groups. As shown in Table 1, children across
the treatment conditions performed in the average range
on measures of fluid intelligence (the Raven Colored
Progressive Matrices Test) and reading (WRAT word read-
ing and TORC reading comprehension). The four MD
treatment conditions were statistically comparable on fluid
intelligence, F(3, 63) = .89, p > .05,
2
= .04; WRAT word
reading, F(3, 64) = 1.38, p > .05,
2
= .06; and TORC com-
prehension, F(3, 63) = .18, p > .05,
2
= .01. No significant
differences across treatment occurred as a function of gen-
der,
2
(4, N = 91) = 3.17, p >.05, or ethnicity,
2
(16, N = 91) =
18.90, p > .05. To reduce the sample-to-task ratio, compos-
ite scores were created to make comparisons between pre-
test and posttest performances. The composite scores for
word problem solving accuracy (CMAT, TOMA-2, KEYM)
and word problem solving components (identification of
question, number, goal, algorithm, operation, and irrelevant
components) were based on the mean z score computed
208 Learning Disability Quarterly 36(4)
from pretest means and standard deviations. Although only
administered at pretest, a composite score was created for
WMC (mean z score of Updating, Conceptual Span, Digit/
Sentence Span).
Pretest Comparisons
A MANOVA was computed among the five treatment con-
ditions on pretest mean z scores for word problem solving
accuracy and identification of word problem solving com-
ponents. As expected, the MANOVA was significant
showing an advantage for children without MD in the
Control condition relative to the conditions that included
children with MD, Wilkss Lambda = .36, F(12, 78) = 8.66,
p < .0001. The univariate tests favored children without
MD in the control condition when compared with other
treatment conditions on the measures of problem-solving
accuracy, F(4, 85) = 32.12, p < .0001,
2
= .60; computa-
tion, F(4, 85) = 6.17, p < .0001,
2
= .22; and identification
of word problem solving components, F(4, 85) = 8.09,
p < .0001,
2
= .28. The post hoc comparisons (Tukey test)
indicated that pretest differences among the treatment con-
ditions that included children with MD were nonsignificant
(all ps > .05) on measures of problem-solving accuracy,
computation, operation span, and identification of word
problem solving components.
Treatment Outcomes
Solution accuracy. The design of this study allowed us
to determine whether the level of posttest performance
achieved by children with MD in the strategy conditions
was statistically comparable to posttest performance in chil-
dren with MD in the control condition. As shown on the left
side of Table 2 for the unconditional model, the random
effect for the intercept between classrooms was significant,
yielding an intraclass correlation of .32 (.319 = [.29 / .29 +
.62]). The fixed effects provided an estimate of the average
intercept for the total sample. The average posttest prob-
lem-solving accuracy z score for Grade 3 children was .63.
Table 2 next shows the first conditional model (Conditional
Model 1) that included treatment conditions without con-
trolling for pretest. Overall, the model indicated that the
mean wave problem solving z score for the participants in
the control condition who were at risk for MD at pretest
was .37. Posttest problem-solving accuracy scores were
.13 (.37 .24), .39 (.37 + .01), and .52 (.37 + .15) for the
Restate, Relevant, and Complete conditions, respectively.
Table 1. Demographic and Psychometric Information for Children by Treatment Conditions
MD Restate MD Relevant MD Complete MD Control NMD Control
Demographics/
psychometrics M SD M SD M SD M SD M SD
Age 8.39 0.50 8.50 0.51 8.22 0.43 8.47 0.52 8.45 0.51
Fluid intelligence and problem solving
RAVEN 105.17 8.12 107.82 10.88 103.88 7.83 104.71 13.94 109.70 10.00
TOMA-2_S 6.89 1.60 6.67 1.24 6.83 1.29 7.33 0.97 9.86 1.74
KEYM_S 6.61 1.38 6.26 1.85 6.94 1.16 6.20 1.62 11.05 1.84
Computation
WIAT 87.72 9.61 88.50 14.28 92.33 9.39 95.67 14.86 103.86 12.77
WRAT-M 96.89 9.54 93.39 9.88 94.47 9.49 94.00 5.79 101.95 10.62
Reading
TORC_S 7.89 1.45 7.35 1.93 7.56 2.66 7.36 1.98 10.10 2.24
WRAT-R 96.67 11.30 100.56 8.64 100.17 10.23 101.36 10.03 110.27 12.55
Working memory
CONCEPT 4.22 6.78
a
3.88 4.29 3.62 2.66 4.43 5.25 7.10 7.71
DIG/SENT 3.72 4.03 3.22 2.53 3.83 3.09 2.86 2.54 4.33 2.72
UPDATE 3.78 3.59 2.82 1.78 3.35 2.87 3.50 2.39 7.09 4.23
Composite
a
.14 .60 .30 .47 .16 .64 .22 .40 .44 .77
% Low WMC
b
66% 72.22% 61.11% 60% 22.73%
Note. MD = children with math difficulties; Restate = Restate Question condition (n = 18, female = 10); Relevant = Restate Question and Isolate
Relevant Number condition (n = 18, female = 6); MD Complete = Restate Question, Isolate Relevant and Irrelevant Number condition (n = 18, female
= 11); MD Control = Control condition with MD (n = 15, female = 7); NMD Control = Control condition without MD (n = 22, female = 11); _S at the
end refers to scale score; RAVEN = Raven Colored Progressive Matrices Test; TOMA-2 = Test of Mathematical AbilitiesSecond Edition; KEYM = KeyMath;
WRAT-M = Wide Range Achievement TestMath Computation subtest; TORC = Test of Reading Comprehension; WRAT-R = Wide Range Achievement Test
Word Reading subtest; CONCEPT = conceptual span; DIG/SENT = Digit sentence span; UPDATE = update task; WMC = working memory capacity.
a
Clearly some children with MD had difficulty on some the isolated working memory tasks; therefore, we performed a square root transformation prior
to creating the composite WM z scores.
b
The binary variable was a median split of the composite scores based on the total sample.
Swanson et al. 209
None of these parameters at posttest were significantly dif-
ferent from the MD Control condition. As expected, a clear
advantage emerged at posttest for children without MD in
the control condition (posttest mean n = 1.43 [.37 + 1.06])
when compared with children with the MD in the control
condition.
The second model (Conditional Model 2) entered the
treatment conditions and grand mean centered pretest scores
for problem-solving accuracy. The results show a signifi-
cant posttest score advantage for children in the Relevant
and Complete conditions when compared with the MD
Control. In addition, the model eliminated differences
between the two control conditions. The model also reduced
the significant random effects related to intercepts between
classrooms when compared with the unconditional means
model and accounted for 69% of the explainable variance
(.29 .09 / .29). The Deviance score was lower than the
unconditional model, as was Akaikes Information Criterion
(AIC is conservative for the number of parameters esti-
mated), suggesting that Conditional Model 2 was a good fit
to the data.
The third conditional model sought to determine whether
entering WMC (binary variable) into the HLM analysis
improved the model fit. This model yielded the lowest devi-
ance and AIC values when compared with the other models.
A comparison of AIC values between Models 2 and 3 indi-
cated that inclusion of WMC provided a better fit to the
data,
2
(1) = 7.30 (157.9 150.6), p < .01. The adjusted
posttest partialed means for children with MD were
.53 (SE = 0.13, unadjusted SD = 0.64), .88 (SE = 0.14,
SD = 0.63), .83 (SE = 0.13, SD = 0.88), .48 (SE = 0.13,
SD = 0.85) for the Restate, Relevant, Complete, and MD
Control conditions, respectively. The adjusted posttest mean
for the control group without MD was .38 (SE = 0.16, SD =
0.79). A Tukey test found a significant difference (all
ps < .05) in adjusted posttest scores in favor of the Relevant
and Complete conditions when compared with the other
conditions (Relevant = Complete > Restate = Control MD =
Control NMD). Although reading comprehension was
entered into Model 3, its influence was nonsignificant (esti-
mate = .009, SE = 0.003, t ratio = .62) and was therefore
removed from the model.
In summary, the results support the notion that para-
phrasing relevant-only (Relevant) and all propositions
(Complete) enhances problem-solving accuracy. In addi-
tion, the results show that WMC is an important moderator
of childrens performance on standardized measures of
problem-solving accuracy. Point-biserial correlations were
computed between the WMC group (binary variable) and
posttest problem-solving accuracy (continuous variable) to
investigate those treatment conditions that created the high-
est demands on WMC. The correlations between WMC and
posttest problem-solving accuracy were significant for the
conditions of Restate, r(16) = .74, p < .001; and Complete,
r(16) = .57, p < .01; but not for Relevant, r(16) = .08,
p > .05; MD Control, r(13) = .36, p > .05; or NMD Control,
r(20) = .17, p > .05. Using Cohens (1988) criteria for high
(r = .50) and moderate effect sizes (r = .30), a high effect
size emerged for the Restate condition and Complete condi-
tion, whereas the lowest demands on WMC were related to
the Relevant and Control conditions.
Identifying problem-solving components. The same model-
ing procedure used with problem-solving accuracy was
computed for correct identification of problem-solving
Table 2. Hierarchical Linear Models Predicting Posttest Solution Accuracy
Unconditional model Model 1 Model 2 Model 3
Fixed effects Estimate SE Estimate SE Estimate SE Estimate SE
Intercept 0.63** 0.09 0.37* 0.18 0.47* 0.13 0.31* 0.13
Restate 0.24 0.25 0.05 0.19 0.06 0.18
Relevant 0.01 0.25 0.38* 0.19 0.40* 0.18
Complete 0.15 0.25 0.36 0.19 0.35* 0.18
Control NMD 1.05** 0.24 0.03 0.21 0.10 0.20
Covariates
Pretest 0.98** 0.12 0.91** 0.10
WMC 0.38** 0.11
Random effects Variance SE Variance SE Variance SE Variance SE

0
2
0.29* 0.18 0.09 0.08 0.04 0.04 0.06 0.05

2
0.62** 0.09 0.51** 0.08 0.26** 0.04 0.25** 0.03
Deviance 231.7 207.55 157.9 150.6
AIC 235.7 211.5 161.9 154.6
Note. NMD = children without math difficulties; WMC = working memory capacity; AIC = Akaikes Information Criterion.
*p < .05. **p < .01.
210 Learning Disability Quarterly 36(4)
components. As shown in Table 3, Model 3 provided the
best fit to the data (yielded the lowest Deviance and AIC
scores). This model eliminated the significant advantage of
the NMD Control group found in Model 1. Furthermore, the
results also showed a significant posttest score advantage
for children with MD in the Complete condition when com-
pared with the control condition that included children with
MD. The adjusted posttest partialed means from Model 2
for children with MD were .32 (SE = 0.09) for the Restate
condition, .56 (SE = 0.10) for the Relevant condition, .69
(SE = 0.09) for the Complete condition, and .42 (SE = 0.10)
for the Control condition that included children with MD.
The adjusted posttest mean for the Control group without
MD was .48 (SE = 0.12). A Tukey test found a significant
difference (all ps < .05) in adjusted posttest scores in favor
of the Complete and Relevant conditions when compared
with the other conditions that included children with MD
(Complete = Relevant, Complete > Restate = Control, Rel-
evant > Restate = Control). When compared with children
without MD, no significant differences (ps > .05) in posttest
scores occurred for children in the Relevant and Complete
conditions (Control NMD = Complete = Relevant > Control
MD = Restate).
The previous results for Model 3 showed that WMC
played a significant role in moderating treatment effects for
identifying problem-solving components. The correlations
between WMC and posttest problem-solving component
identification were not significant for the Restate, r(16) = .30,
p > .05; Relevant, r(16) = .30, p > .05; Complete, r(16) = .32,
p > .05; or MD Control, r(13) = .34, p > .05, conditions, but
they were significant for the NMD Control condition,
r(20) = .45, p < .05. Using Cohens criteria, moderate effect
sizes in favor of children with higher WMC occurred across
all conditions.
Operation span. The same modeling procedure used
with problem-solving components was computed for the
operation span item recall measure. As shown on the right
side of Table 4, Model 2 provided the most parsimonious
model (yielded the lowest AIC scores), and Model 3,
which entered WMC, provided a poor fit. The results also
showed a significant posttest score advantage for children
with MD in the Restate and Relevant conditions when
compared with the MD Control condition. The adjusted
posttest means from Model 2 for children with MD were
.80 (SE = 0.31) for the Restate condition, .83 (SE = 0.32)
for the Relevant condition, .34 (SE = 0.30) for the Com-
plete condition, and .20 (SE = 0.35) for the MD Control
condition. The adjusted posttest mean for the Control
group without MD was .17 (SE = 0.37). A Tukey test
found a significant difference (all ps < .05) in adjusted
posttest scores in favor of the Restate and Relevant condi-
tions when compared with the other conditions (Restate =
Relevant > Complete = Control MD = Control NMD). In
summary, the generative strategy conditions that included
restating the question and attending to relevant proposi-
tion increased operation span performance relative to the
other conditions. The correlations between WMC pretest
and posttest operation span were not significant for the
Restate, r(16) = .17, p > .05; Relevant, r(16) = .31, p > .05;
Complete, r(16) = .38, p > .05; MD Control, r(13) = .20,
p > .05; or NMD Control, r(20) = .04, p > .05, conditions.
Using Cohens criteria, moderate effect sizes in favor of
Table 3. Hierarchical Linear Models Predicting Posttest Identification of Problem Components
Unconditional
model Model 1 Model 2 Model 3
Fixed effects Estimate SE Estimate SE Estimate SE Estimate SE
Intercept 0.50** 0.04 0.40* 0.11 0.41** 0.10 0.31* 0.11
Restate 0.18 0.15 0.11 0.14 0.10 0.13
Relevant 0.06 0.15 0.14 0.14 0.14 0.14
Complete 0.19 0.15 0.27* 0.14 0.27* 0.13
Control NMD 0.43** 0.24 0.10 0.16 0.06 0.15
Covariates
Pretest 0.21** 0.07 0.18** 0.08
Problem-solving accuracy 0.19** 0.16 0.16** 0.08
WMC 0.24** 0.09
Random effects Variance SE Variance SE Variance SE Variance SE

0
2
0.03 0.03 0.003 0.01 0.0 0.04 0.002 0.01

2
0.19** 0.03 0.16** 0.02 0.14** 0.04 0.13** 0.02
Deviance 116.7 104.4 94.5 90.0
AIC 120.7 108.4 96.5 94.0
Note. NMD = children without math difficulties; WMC = working memory capacity; AIC = Akaikes Information Criterion.
*p < .05. **p < .01.
Swanson et al. 211
children with higher WMC occurred for only the Relevant
and Complete conditions.
Discussion
This study investigated whether generative strategies that
focused on paraphrasing propositions within word problems
facilitated posttest problem-solving accuracy, identification
of problem-solving propositions (problem-solving compo-
nents), and operation span performance in children with MD.
Overall, the results show that generative treatment conditions
that focused on relevant-only propositions or all propositions
facilitated posttest solution accuracy performance when
compared with the control conditions that included children
with and without MD. A performance advantage was found
for the Complete generative condition for posttest scores
related to proposition identification, whereas a posttest
advantage was found on operation span measures for treat-
ments that included a restatement of the question proposi-
tion and/or relevant propositions. The results will now be
discussed in terms of three questions that directed the study.
Research Question 1: Do generative strategy con-
ditions that focus on isolated propositions or all
propositions within word problems facilitate solu-
tion accuracy when compared with the control
condition for children with MD?
When compared with the Control condition, the results
showed an advantage at posttest for treatment conditions
that either focused on relevant-only propositions (Relevant
condition) or all the propositions (Complete condition). Both
conditions significantly predicted posttest solution accuracy
when compared with the Control conditions. It is important
to note, however, that a performance advantage only occurred
for the Complete condition when WMC was entered into the
equation. Thus, without covarying WMC in the analysis, no
significant advantage was found for the Complete generative
training condition. The results also showed no significant
advantages relative to the Control condition for merely para-
phrasing propositions related to the goal/questions within
word problems (Restate condition).
Overall, the findings showed that generative strategy
training that included relevant-only propositions or all
propositions facilitated solution accuracy. The results are
consistent with the idea that knowing how to recognize and
represent relational statements is significantly correlated
with a students problem-solving performance (Mayer,
Lewis, & Hegarty, 1992) as well as WM skill (Swanson
et al., 2008). However, we did not find support for the notion
that writing instructions directed toward specific proposi-
tional phrases (i.e., identifying the question or identifying
relevant-only propositions) would yield higher outcomes on
problem-solving accuracy than the contrast (Complete)
condition.
Research Question 2: Do generative strategies make
demands on WMC?
Although an answer to this question is in the affirmative,
there are alternative explanations for findings. We consider
four possibilities. One hypothesis is that because the read-
ing, computation, and general fluid intelligence for children
with MD were intact, the reliable use of generative strategies
Table 4. Hierarchical Linear Models Predicting Posttest Operation Span
Unconditional
model Model 1 Model 2 Model 3
Fixed effects Estimate SE Estimate SE Estimate SE Estimate SE
Intercept 0.30* 0.13 0.23 0.34 0.21 0.35 0.25 0.37
Restate 0.74 0.44 0.98* 0.46 0.98* 0.46
Relevant 0.72 0.44 1.02* 0.46 1.02* 0.46
Complete 0.38 0.44 0.54 0.45 0.52 0.46
Control NMD 0.69 0.43 0.02 0.53 0.01 0.54
Covariates
Pretest 0.03 0.13 0.03 0.13
Problem-solving accuracy 0.59* 0.26 0.57* 0.26
WMC 0.12 0.28
Random effects Variance SE Variance SE Variance SE Variance SE

0
2
0.003 0.07 0.02 0.08 0.07 0.11 0.06 0.11

2
1.50** 0.23 1.48** 0.23 1.42** 0.23 1.44** 0.23
Deviance 290.1 286.9 279.5 280.0
AIC 294.1 290.9 283.5 284.0
Note. NMD = children without math difficulties; WMC = working memory capacity; AIC = Akaikes Information Criterion.
*p < .05. **p < .01.
212 Learning Disability Quarterly 36(4)
superseded the role that any individual differences in WMC
might play. In contrast to this hypothesis, however, we found
that when WMC was entered into HLM analysis, a better fit
to the data was found when predicting problem-solving
accuracy. Thus, the results do not support the concept that
WMC plays a secondary role in problem-solving outcomes
related to treatment conditions for children with MD.
Another possible explanation is that a limited-capacity
WM system underlies word problem solving difficulties in
children with MD. This hypothesis assumes that although
WMC may act in tandem with other processes, this general
system may operate independent of strategy conditions.
The present study clearly showed that WMC accounted for
important variance in predicting problem-solving accuracy;
however, significant correlations between WMC and post-
test solution accuracy were isolated to specific generative
conditions. High effect sizes occurred in correlations
between WMC and posttest problem-solving accuracy for
the Restate (r = .74) and Complete (r = .57) conditions.
Another possible explanation is that generative train-
ing compensates for individual differences in WMC. Some
studies have shown that strategy training helps low-span
participants allocate WM resources more efficiently when
compared with high-span participants (e.g., Turley-Ames &
Whitfield, 2003). Thus, we expected that children with MD,
especially those with relatively lower WM span, would
benefit more from generative instruction when compared
with the Control condition. Such was not the case in this
study. The results related to the correlations within treat-
ment condition showed positive correlations between WMC
and posttest outcomes.
The final model, the one we prefer, suggested that gen-
erative training was more likely to improve problem-solv-
ing outcomes for children with a relatively larger WMC
because these children have spare WM sources to effec-
tively use these strategies. The general patterns of the cur-
rent study are in line with this model. High effect sizes
(positive correlations) between WMC and posttest prob-
lem-solving accuracy occurred for the Restate (r = .74) and
Complete (r = .57) conditions, but weak or moderate effect
sizes were found for the Relevant (r = .08), MD Control
(9r = .36), and NMD Control (r = .17, p > .05) conditions.
Thus, the children who benefited most from the Restate and
Complete conditions had relatively higher WMC.
Research Question 3: Does generative strategy
instruction facilitate transfer?
The results suggest that solution accuracy for children
with MD improved as a function of generative training. The
results also showed positive near-transfer effects related to
identifying problem-solving components and far transfer
related to the operation span measure. The results related to
component identification would be expected because the
training sessions directed children to such components.
Thus, the results on the operation span measure were of
more interest. The results show that adjusted posttest
operation scores were higher for the Restate and Relevant
conditions when compared with the Control and Complete
conditions. No studies we are aware of, however, have
shown that strategy training within an academic domain
(word problem solving) directly influences WM or vice
versa (e.g., see the Holmes, Gathercole, & Dunning, 2009,
discussion of the sleeper effect). Perhaps the approach we
took to enhance transfer by embedding WM demands
(increasing sentence load by increasing the number of irrel-
evant propositions) within the curriculum may be an impor-
tant avenue in future research.
Summary
The results support the use of generative strategies for
improving problem-solving performance in children with
MD. Contingent on the outcome measure, generative strat-
egy conditions significantly boosted performance at post-
test. The results also suggest that some generative conditions
placed greater WM demands on children with MD than
other conditions.
Acknowledgments
The authors are indebted to Heather Owen, Ryan Grimm, Danielle
Guzman-Orth, Melina Melgarejo, and Quinn Tomlinson for their
assistance in the data collection. Appreciation is given to the Santa
Barbara and Goleta School Districts for their participation.
Special appreciation is given to the undergraduate students, teach-
ers, and administrators who made this study possible.
Authors Note
The study does not necessarily reflect the views of the U.S.
Department of Education or the school district.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect
to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support
for the research, authorship, and/or publication of this article: This
article is based on a study funded by the U.S. Department of
Education, Cognition and Student Learning (R324A09002),
Institute of Education Sciences, awarded to H. Lee Swanson.
Note
1. Reviewers raised concerns related to sample size, specifically
concerns were raised related to power in our analysis and our
standard errors (SE, that is, the larger the sample size the
smaller the SE; see Maas & Hox, 2005). According to Maas
and Hox (2005), at least 50 Level 2 observations (classrooms
Swanson et al. 213
in this case) are needed to assure that SE estimates for the fixed
effect components are unbiased. Furthermore, increasing the
sample size at Level 1 does nothing to enhance power at Level 2
(Bickel, 2007). Thus, because significant results were found at
Level 1 and the random effects for the intercepts were not
significant in the best fitting model, the data provide an ade-
quate test of our hypotheses. To partially address the sample
size issue, however, effect sizes (ESs) were computed. Our
hierarchical linear model (HLM) analyses adjusted for the
covariates of pretest and working memory capacity (WMC),
and therefore when estimating ESs, we accounted for the
dependence among measures within the same classroom. We
calculated Hedges g = /[(SE
1
2
)(N
1
)+(SE
2
2
)(N
2
)/2]

, where
is the HLM coefficient for the adjusted posttest mean differ-
ence between treatment (partial for pretest and working mem-
ory), adjusted for both Level 1 and Level 2 covariates, and N
1

and N
2
were the sample sizes. SE
1
and SE
2
were the SE for the
posttest treatment conditions, respectively. The Level 2 coef-
ficients were adjusted for the Level 1 covariates such that the
Level 1 covariates (pretest and WMC) were grand mean cen-
tered. For the interpretation of the magnitude of the ESs,
Cohens (1988) distinction was used; an ES of 0.20 is consid-
ered small, and ESs of 0.50 and 0.80 are considered moderate
and large, respectively. For the purposes of brevity, only ESs
between the control MD condition and remaining conditions
are reported. Moderate ESs are in bold. Positive ESs are in
favor of the treatment condition relative to the control condi-
tion that included children with MD. For problem-solving
accuracy, the ESs for the control MD condition when com-
pared with the restate, relevant, complete, and non-MD control
condition were .09, .70, .64, and .15, respectively. For prob-
lem-solving processes, the ESs for the control MD condition
when compared with the restate, relevant, complete, and non-
MD control condition were .24, .33, .67, and .12, respec-
tively. For operation span, the ESs for the control MD
condition when compared with the restate, relevant, complete,
and non-MD control condition were .71, .72, .39, and .01,
respectively. It is important to note that miniscule ESs
emerged between MD and NMD groups in the control condi-
tion reflects the fact that pretest and WMC partialed out the
difference in performance but does not imply equivalence in
problem-solving accuracy.
References
Altier, M. B., Day, R., & Balka, D. (2009). Macmillan/McGraw-
Hills (2009) Mathematics: Concepts, skills, and problem solving
Third grade. Columbia, OH: Macmillan/McGraw-Hill.
Baker, S., Gersten, R., & Lee, D. (2002). A synthesis of empirical
research on teaching mathematics to low-achieving students.
Elementary School Journal, 103, 5173.
Bickel, R. (2007). Multilevel analysis for applied research: Its
just regression. New York, NY: Guilford.
Brown, V. L., Cronin, M. E., & McIntire, E. (1994). Test of math-
ematical ability. Austin, TX: PRO-ED.
Brown, V. L., Hammill, D., & Weiderholt, L. (1995). Test of read-
ing comprehension. Austin, TX: PRO-ED.
Bryk, A. S., & Raudenbush, S. W. (2002). Hierarchical linear
models: Applications and data analysis methods. London,
England: SAGE.
Cohen, J. (1988). Statistical power analysis for the behavioral sci-
ences (2nd ed.). Hillsdale, NJ: Lawrence Erlbaum.
Connolly, A. J. (1998). KeyMath revised/normative update. Circle
Pines, MN: American Guidance Services.
Cook, J. L., & Rieser, J. J. (2005). Finding the critical facts: Chil-
drens visual scan patterns when solving story problems that
contain irrelevant information. Journal of Educational Psy-
chology, 97, 224234.
Cornoldi, C., Drusi, S., Tencati, C., Giofr, D., & Mirandola, C.
(2012). Problem solving and working memory updating diffi-
culties in a group of poor comprehenders. Journal of Cognitive
Education and Psychology, 11, 3944.
Fletcher, J. M., Epsy, K. A., Francis, P. J., Davidson, K. C.,
Rourke, B. P., & Shaywitz, S. E. (1989). Comparison of cutoff
and regression-based definitions of reading disabilities. Jour-
nal of Learning Disabilities, 22, 334338.
Geary, D. C. (2010). Mathematical disabilities: Reflections on cog-
nitive, neuropsychological, and genetic components. Learning
and Individual Differences, 20, 130133.
Gersten, R., Chard, D. J., Jayanthi, M., Baker, S. K., Morphy, P.,
& Flojo, J. (2009). Mathematics instruction for students with
learning disabilities: A meta-analysis of instructional compo-
nents. Review of Educational Research, 79, 12021242.
Holmes, J., Gathercole, S. E., & Dunning, D. L. (2009). Adap-
tive training leads to sustained enhancement of poor working
memory in children. Developmental Science, 12, 915.
Hooper, S., Sales, G., & Rysavy, S. D. M. (1994). Generating
summaries and analogies alone and in pairs. Contemporary
Educational Psychology, 19, 5362.
Hresko, W., Schlieve, P. L., Herron, S. R., Swain, C., & Sherb-
enou, R. (2003). Comprehensive math abilities test. Austin,
TX: PRO-ED.
Kail, R., & Hall, L. K. (1999). Sources of developmental change
in childrens word-problem performance. Journal of Educa-
tional Psychology, 91, 660668.
Kolloffel, B., Eysink, T., de Jong, T., & Wilhelm, P. (2009). The
effects of representational format on learning combinatorics
from an interactive computer simulation. Instructional Sci-
ence, 37, 503517.
Maas, C., & Hox, J. (2005). Sufficient sample sizes in multiple
regression analysis. Methodology, 1, 8692.
Mayer, R. E., Lewis, A. B., & Hegarty, M. (1992). Mathematical
misunderstandings: Qualitative reasoning about quantitative
problems. In J. I. D. Campbell (Ed.), The nature and origins of
mathematical skills (pp. 137154). Amsterdam, Netherlands:
Elsevier.
Montague, M. (2008). Self-regulation strategies to improve math-
ematical problem solving for students with learning disabili-
ties. Learning Disability Quarterly, 31, 3744.
214 Learning Disability Quarterly 36(4)
Montague, M., Warger, C., & Morgan, T. H. (2000). Solve it!
Strategy instruction to improve mathematical problem solv-
ing. Learning Disabilities Research & Practice, 15, 110116.
Passolunghi, M. C., Marzocchi, G. M., & Fiorillo, F. (2005).
Selective effect of inhibition of literal or numerical irrelevant
information in children with attention deficit hyperactiv-
ity disorder (ADHD) or arithmetic learning disorder (ALD).
Developmental Neuropsychology, 28, 731753.
Plass, J. L., Chun, D. M., Mayer, R. E., & Leutner, D. (1998).
Supporting visual and verbal learning preferences in a second-
language multimedia learning environment. Journal of Educa-
tional Psychology, 90, 2536.
Powell, S. R. (2011). Solving word problems using schemas: A
review of literature. Learning Disabilities Research & Prac-
tice, 26, 94108.
Psychological Corporation. (1992). Wechsler Individual Achieve-
ment Test. San Antonio, TX: Harcourt Brace.
Raven, J. C. (1976). Colored progressive matrices test. London,
England: H. K. Lewis.
SAS Institute. (2010). SAS/STAT users guide. Cary, NC: Author.
Schwamborn, A., Mayer, R. E., Thillmann, H., Leopold, C., &
Leutner, D. (2010). Drawing as a generative activity and draw-
ing as a prognostic activity. Journal of Educational Psychol-
ogy, 102, 872879.
Siegel, L. S., & Ryan, E. B. (1989). The development of working
memory in normally achieving and subtypes of learning dis-
abled. Child Development, 60, 973980.
Swanson, H. L., & Beebe-Frankenberger, M. (2004). The relation-
ship between working memory and mathematical problem
solving in children at risk and not a risk for serious math dif-
ficulties. Journal of Educational Psychology, 96, 471491.
Swanson, H. L., Cooney, J. B., & Brock, S. (1993). The influence
of working memory and classification ability on childrens
word problem solution. Journal of Experimental Child Psy-
chology, 55, 374395.
Swanson, H. L., Jerman, O., & Zheng, X. (2008). Growth in work-
ing memory and mathematical problem solving in children at
risk and not at risk for serious math difficulties. Journal of
Educational Psychology, 100, 343379.
Swanson, H. L., Kehler, P., & Jerman, O. (2010). Working mem-
ory, strategy knowledge, and strategy instruction in children
with reading disabilities. Journal of Learning Disabilities, 43,
2447.
Turley-Ames, K. J., & Whitfield, M. (2003). Strategy training and
working memory performance. Journal of Memory and Lan-
guage, 49, 446468.
van Garderen, D. (2007). Teaching students with LD to use dia-
grams to solve mathematical word problems. Journal of
Learning Disabilities, 40, 540553.
Vilenius-Tuohimaa, P. M., Aunola, K., & Nurmi, J. (2008). The
association between mathematical word problems and reading
comprehension. Educational Psychology, 28, 409426.
Wilkinson, G. S. (1993). Wide Range Achievement Test. Wilmington,
DE: Wide Range.
Wittrock, M. C. (1989). Generative processes of comprehension.
Educational Psychologist, 24, 345376.
Wittrock, M. C. (1991). Generative teaching of comprehension.
Elementary School Journal, 92, 169184.
Xin, Y. P. (2008). The effect of schema-based instruction in solv-
ing mathematics word problems: An emphasis on prealge-
braic conceptualization of multiplicative relations. Journal for
Research in Mathematics Education, 39, 526551.

You might also like